[LuisFigueroaG]

With skip_records_with_error, two errors don't behave like the others: CSV_MAX_RECORD_SIZE and CSV_NON_TRIMABLE_CHAR_AFTER_CLOSING_QUOTE are returned straight from the parse loop (return this.__error(...)). When the record is skipped, __error returns undefined, so parse just stops in the middle of the buffer without saving the rest in previousBuf.

The result is a skip event followed by silent data loss:

import { parse } from "csv-parse/sync";

parse('"a" b,c\nd,e\nf,g\n', { rtrim: true, skip_records_with_error: true });
// []  (expected [["d", "e"], ["f", "g"]])

With the stream API it can also produce wrong records, because the next chunk is concatenated with the stale previousBuf:

const parser = parse({ max_record_size: 5, skip_records_with_error: true });
parser.write("x,y\nab");
parser.write("cdefghij,c\nd,e\n");
parser.write("f,g\n");
parser.end();
// [["x", "y"], ["abf", "g"]]

This checks the result of __error like the other skippable errors, so the record is flagged and dropped in __onRecord and parsing goes on. Both checks are skipped once the record already has an error, otherwise every following character of the record would emit another skip event. Without skip_records_with_error nothing changes, since recordHasError is only set in that mode.

Tests are added to test/option.skip_records_with_error.ts for both errors, the max_record_size one with the oversized field split across writes. Both fail on master.

Validation (Ubuntu, Node 24.22.0):

• npm test in packages/csv-parse: 631 passing, 2 pending
• npm run lint:check: no errors (5 existing pending-test warnings)
• prettier check on the changed files